Bivariate Error Correction Model (ECM) with GARCH error

[johnsirius]

Dear sir,

I m trying to use ECM-GARCH to estimate conditional hedge raio using eviews 6, i ve been trying to do this for 3 days but still couldnt get the result :(

The dynamic hedge ratio at time t can be computed as the ratio of conditional covariance between spot and future to the conditional variance of future.

But i cant find the conditional covariance and conditinal variance for my model, and so frustrated at the moment.

I only know how to run GARCH on a Univariate system not on a bivariate vector error correction model (VECM), can you please help me?? :(

Thanks so much!!!

John

[AHS]

Hi,

I am currently experiencing some of the same problems as you have encountered.

I have found a description of how to estimate a multivariate VAR-GARCH in Brooks, C., 2008. Introductory Econometrics for Finance (2nd ed). Cambridge University Press. pp.441-444. This is done by highlighting the variables you want to include, right-clicking, and selecting Open -> As System. Choose the appropriate specification and click OK. Then click Proc -> Estimate... and choose Estimation method ARCH. Click OK and I think you get the VAR-GARCH. In your case click view and conditional covariance. The problem is however that this doesn't work for a VECM. In my case another problem is that the method does not allow me to run Granger Causality tests/impluse response analysis (available when estimating a regular VAR or VECM).

What I am trying to do is to investigate the lead-lag relationship between spot and futures prices (Granger Causality tests and Impulse Response Analysis).
I have three pairs of series and after employing the Johansen test I have found that the first pair have rank=1, while the other two have rank=2. When estimating "normal" VECM (1st pair) and VAR (pair 2 and 3) models I get clear indications of heteroskedasticity in the residuals. To my knowledge I have two possibilities when trying to remedy this;

1. After studying some articles I have found that Lee's (1994) VECM-GARCH-X specification should be employed for the first pair, while a VAR-GARCH-BEKK is sufficient for the other two pairs. However, I do not know how to estimate the VECM-GARCH-X or conduct the Granger Causality tests/impulse response analysis for any of the pairs.

2. Another way to solve the problem would be to use White's heteroskedasticity consistent variance-covariance matrix, but I do not know how this works or how to incorporate it for a VAR or VECM.

I have a few other statistical packages than EViews on my computer, so if I have to use another program do not hesitate to tell me.

AHS

[johnsirius]

Hi AHS,

Thanks a lot for your helpful tips!!

and with your question, Im not up to that part so wont be able to help you much about it, but good luck on getting mor responds.

However, I did see the option click when coming across the VECM-GARCH, im not sure if this would help but try this:

Open your VECM for 1 out of your 2 pairs, ( believe its in VAR system) go to Proc > Make System > select "order by variable/lag" > click on Proc again > Change estimation method to ARCH > Change ARCH model specification to Diagnoal BEKK ...and specify the rest of the options based on your case.

I hope this will give you a bit of hint :D (SO, by doing this, you have actually incorporated the VECM into the GARCH estimation in EVIEWS)?

Best,
John

[spencerprosser]

This was amazing whoever figured this out but has anyone figured out how to get a dummy variable into the varcov matrix... i know how to put a variable into the varcov matrix but it will not let me multiply it by the varcov matrix (M) it gives me B*dummy and i have no idea how to change what it is doing. Anyideas? my goal would be show how the varcov matrix has changed over periods.
Regards,